Active learning for phenotyping tasks

Abstract:

Active learning is a popular research area in machine learning and general domain natural language processing (NLP) communities. However, its applications to the clinical domain have been studied very little and no work has been done on using active learning for phenotyping tasks. In this paper we experiment with a specific kind of active learning known as uncertainty sampling in the context of four phenotyping tasks. We demonstrate that it can lead to drastic reductions in the amount of manual labeling when compared to its passive counterpart.